In one pilot, a routine optimized per tool advice now runs faster but occasionally returns stale data. What is the most likely explanation?

The correct explanation is that the optimization was accepted without validating that it preserved the routine's original behaviour, and it altered how or when data is refreshed. Optimizations that change caching, batching or timing can shift when data is refreshed even while the code runs faster overall, which is exactly the kind of behavioural side effect validation is meant to catch before acceptance. The claim that stale data is unrelated to the optimization and is a pre-existing issue that coincidentally surfaced afterwards is incorrect because dismissing the timing as coincidence skips the investigation the symptom calls for. The claim that the optimization tool's suggestions cannot affect data freshness, only execution speed, is incorrect because optimization changes routinely affect timing-sensitive behaviour such as when data is read or cached, not only raw execution speed. The claim that the pilot environment's data source is outdated and needs to be refreshed independently of the code change is incorrect because it redirects attention to the environment instead of examining the change that was actually made to the routine. For the exam, treat any new symptom appearing after an optimization as something the optimization should be checked against first.